The Article 50 checklist for AI startups
Twelve concrete steps to get a generative AI product ready for the EU AI Act's transparency rules, from working out your role to keeping evidence.
6 min read · Updated 27 October 2026 · Markproof
In short
- Article 50 of the EU AI Act has applied since 2 August 2026. If your generative AI product was on the market before then, the marking duty applies from 2 December 2026.
- Most of the work for a generative AI startup is marking outputs (50(2)), plus disclosure duties that depend on your product.
- Work through the steps below, and keep the evidence as you go.
The checklist
Work out your role for each AI feature
If you offer an AI system under your own name, you're likely a provider, even if the model comes from someone else. If you only use AI tools, you're a deployer. Many startups are both. More on roles.List every type of content you generate
Images, audio, video and text each need marking. Note which features produce which, and where the files end up.Check whether an exception applies
Purely assistive editing, and features that don't substantially alter the user's input, are exempt from the marking duty. A prompt-to-image generator is not.Add machine-readable marking at the point of generation
Mark every output before it reaches users. Use two layers, as the EU Code of Practice expects: a signed C2PA manifest and an invisible watermark.Make sure the marks survive your own pipeline
Converting, resizing or re-saving files after marking can strip the C2PA manifest. Mark as the final step, and serve the signed file as-is.Test robustness on your real outputs
Check that marks survive what usually happens to your files: JPEG re-compression, resizing, cropping, screenshots. Article 50 asks for marking that's robust and reliable as far as technically feasible.Provide a way to detect the marks
Marked content must be detectable. Offer a verification endpoint or a public checker that deployers, users and authorities can use.Disclose chatbots and other interactive AI
Under 50(1), people must know they're interacting with an AI system, unless it's obvious. Show this clearly, at the latest at the first interaction.Label deepfakes and public-interest text, if you publish them
As a deployer under 50(4), disclose AI-generated or manipulated media that looks real, and AI-written text published to inform the public, unless it went through human editorial review. The Code of Practice includes guidance on label placement and an EU icon.Review the Code of Practice and decide whether to sign
The Code is voluntary but sets the practical bar. If you don't sign it, be ready to explain how your own measures meet the same requirements.Check your vendors and contracts
If a model or API provider handles part of the marking, confirm what they cover, in which formats, and get it in writing.Keep evidence
Record your role assessment, marking methods, test results and when marking went live. If a regulator asks, documentation is what shows good faith. (Markproof's Pro plan generates a technical document for this.)
Key dates
2 August 2026: Article 50 applies, and new generative AI systems must mark outputs from launch. 2 December 2026: the marking grace period ends for systems already on the market before 2 August 2026.
Breaches can lead to fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs and start-ups, whichever is lower.
Sources
This guide is general information, not legal advice. Laws and guidance change; check the sources, and talk to a qualified lawyer about your situation.
Check your own output in seconds.
Drop a file into the free checker to see whether it carries a watermark and Content Credentials. Then mark your files with one API call: 250 free calls a month, no card required.